The AI graveyard: a running list of projects and startups that didn’t make

The artificial intelligence sector is undergoing an aggressive shakeout, marked by the shuttering of standalone products, failed hardware devices, and internal corporate retrenchments documented in TechCrunch's report. S&P Global Market Intelligence estimates that approximately 42% of AI initiatives are ultimately abandoned by their corporate parent organizations due to insufficient capital, engineering roadblocks, intense platform competition, and weak user demand. Even well-funded ventures and hyperscale tech giants are now actively terminating or scaling back products that fail to retain paying users.

This retrospective details the specific startups, specialized apps, and enterprise platform features that recently ceased operations or experienced severe public setbacks. Engineering leaders, product designers, and technical founders can study these casualty reports to understand how rapidly commoditizing foundation models and native platform integrations consistently undermine fragile software wrappers and overhyped companion gadgets.

AI graveyard: what actually changed

The casualty list encompasses both independent startups and core initiatives within major tech firms. Independent workflow automation tool Relay shut down in August after five years in business because frontier model providers began baking native agentic triggers directly into foundational products. Consumer chatbot platform Character.ai filed for bankruptcy five years after its inception, while role-play narrative hub Figgs AI shut down after finding it too expensive to provide free hosting for its one million users. Standalone comparative hub Yupp, which allowed users to benchmark more than 800 models side by side from providers like OpenAI, Google, and Anthropic while earning cryptocurrency, folded in March 2026 after failing to establish viable product-market fit.

Corporate tech giants encountered severe friction when launching dedicated standalone tools. OpenAI ended ChatGPT Atlas, its dedicated AI-driven web browser, on August 9 after less than a year on the market, while absorbing its core capabilities back into ChatGPT. OpenAI also wound down its standalone web agent Operator, curtailed the standalone footprint of DALL-E in favor of native ChatGPT generation, and shuttered its video-sharing platform Sora in March 2026 due to steep inference costs and low user retention. Meanwhile, Notion scheduled the shutdown of its dedicated Notion Mail inbox client for September 22 after noticing customers preferred using external agents to organize communications.

AI graveyard: how it works

Platform consolidations follow a distinct operational pattern where standalone products get absorbed or dismantled when native foundation layers evolve. OpenAI attempted to transform ChatGPT into an enterprise super app on July 9 by deploying dedicated Chat, Codex, and Work environments while relabeling the standard interface ChatGPT Classic. The dense configuration created immediate interface friction, compelling OpenAI to quickly roll back the design changes and restore the familiar single-window experience. Features previously separated across isolated applications, such as browser navigation, autonomous execution, and image synthesis, are now channeled directly through standard conversational prompts.

Hardware experiments faced deeper operational barriers that software updates could not easily resolve. Humane raised $230 million to deliver screenless ambient computing through its wearable AI Pin, but the unit suffered from severe latency issues, thermal throttling, and a battery fire hazard in its charging case. Humane ceased operations in February 2025 before selling the company assets to HP for $116 million. Rabbit sold 100,000 units of its R1 pocket companion after unveiling it at CES in January 2024, yet users reported unfinished software and sparse integrations. Rabbit is now reorienting the R1 toward agentic desktop computer control and developing Project Cyberdeck, a dedicated portable terminal aimed at vibe-coding workflows.

AI graveyard: why it matters now

The ongoing closures demonstrate that specialized application wrappers face existential obsolescence as major platforms bundle equivalent capabilities into core offerings. Former developers of Google's NotebookLM launched an AI audio utility called Huxe to generate podcast-style discussions from written source files, but the startup ceased operations in May 2026 as broad-distribution audio platforms like Spotify expanded native synthetic voice generation tools. When foundational platforms supply standard models and native interfaces at scale, niche startups that rely entirely on underlying application programming interfaces cannot sustain high infrastructure overhead alongside low switching costs.

Enterprise platform rollouts also face heightened regulatory, privacy, and performance scrutiny that disrupts release roadmaps. Apple unveiled Apple Intelligence in 2024 with promises of an overhauled, context-aware Siri, but software defects and engineering bottlenecks caused prolonged delays that ultimately led to a $250 million legal settlement concerning iPhone 16 marketing claims. Apple only introduced the revised Siri to iOS 27 beta testers in July, initiating an English-language rollout this month. Similarly, Microsoft delayed Windows Recall for almost a year following fierce backlash over its continuous screenshot indexing feature, which cybersecurity researchers subsequently showed remained vulnerable to local extraction exploits.

AI graveyard: who is affected

Software engineers, consumer subscribers, and enterprise procurement teams bear the primary impact of these service terminations. Early purchasers of the Humane AI Pin lost hardware support when operations halted and assets transferred to HP, while early adopters of Notion Mail must transition back to standard email workflows ahead of the September 22 service termination. Users relying on free services, such as the one million role-players on Figgs AI or benchmark testers on Yupp, lost access to their historical interactions and community environments entirely when infrastructure bills outpaced platform monetization.

Organizations investing in enterprise workflow automation must now evaluate whether their software partners can endure sustained competition from platform providers. Relay's five-year run ended because native workspace automations inside foundation models eliminated the operational need for an external middle layer. Enterprise teams deploying Windows Copilot systems must also navigate persistent administrative security reviews surrounding Windows Recall, weighing productivity features against internal compliance standards regarding data capture and endpoint vulnerability.

AI graveyard: what to watch

The market will reveal whether standalone AI hardware form factors can outlast general-purpose smartphones. Rabbit's transition toward agentic operating system control on the R1 and its pursuit of the Project Cyberdeck hardware platform will serve as an indicator for whether dedicated physical devices can discover sustainable utility. Observers should track whether specialized agent controllers can maintain sufficient reliability across diverse operating system environments, or if consumers will consistently demand that agentic tasks execute inside existing mobile and desktop operating systems.

On the software front, monitoring will focus on whether Apple's updated Siri satisfies expectations as the feature expands beyond its initial English release into international territories. Concurrently, enterprise buyers will watch how Microsoft handles lingering security vulnerabilities in Windows Recall data storage as the feature moves through wider commercial deployments. As foundation model builders continue integrating multimodal capabilities and agentic execution directly into unified subscription tiers, specialized tooling layers will remain under continuous economic pressure to justify independent subscription pricing.